Research On Block Adaptive Weight Face Recognition System Based On SIFT

The technology of face recognition has made great progress in recent years. Perfect identification effects have been achieved by some classical face recognition methods under ideal conditions. However, in the non-ideal cases, the changes of expressions, illumination, posture changes, and the partial occlusions conditions, recognition rate declined obviously and cannot achieve ideal results. Aiming at this problem, based on the basis of the research in the domestic and overseas face recognition system, this paper presents facial features extraction algorithms which use algorithm SIFT operator which has the advantage of good recognition under non-ideal conditions of face recognition. The key of the research is how to extract SIFT face image feature that is more robust to the facial expression, illumination variations and the partial occlusions, and try to apply it to face recognition system in order to improve the face recognition rate.In this paper, in order to improve the face recognition rate, in view of the traditional image block feature extraction method without considering the block edge information and without considering the problem of the block contribute for the whole. This paper proposed an improved block adaptive weighted face feature extraction method based on SIFT algorithm and applied to face recognition; Firstly, face image is divided into blocks by using block method which is proposed in this paper. Secondly, extracting SIFT feature vector of each sub-block; Then, extracted feature vector is reduced dimension by 2DICA algorithm and we make adaptive weight for feature description of reduction dimension. Later, the paper make use of face image feature extraction and neural network recognition algorithm to make classification recognition for face feature. Finally combining with the experiment, we make the verification and comparative analysis of this method for the recognition rate.In order to do face recognition experiments in face recognition system using feature extraction method of this paper, we choose to test in YALE, ORL and YALEB face repository, as well as trial cooperation in expression changes, illumination changes, position change, shade change with several local characteristics description method. It is proved that this method have higher recognition rate under the condition of shedding, pose, illumination and expression changes compared with the other description method.